If executing from a C or C++ based program. Ctrl-C should do the right thing, and terminate the local program. If the co-simulation is running across multiple machines then the remaining programs won’t terminate properly and will either timeout or if that was disabled potentially deadlock.
Signal handler facilities¶
The C shared library has some facilities to enable a signal handler.
/** Load a signal handler that handles Ctrl-C and shuts down all HELICS brokers, cores, and federates then exits the process.*/ void helicsLoadSignalHandler(); /** clear HELICS based signal Handlers*/ void helicsClearSignalHandler(); /** Load a signal handler that handles Ctrl-C and shuts down all HELICS brokers, cores, and federates then exits the process. This operation will execute in a newly created and detached thread returning control back to the calling program before completing operations.*/ void helicsLoadThreadedSignalHandler();
This function will insert a signal handler that generates a global error on known objects and waits a certain amount time, clears the print buffer, and terminates.
There are also threaded versions of these signal handlers. These generate a separate thread and detach it before executing the HELICS callback operations. This is useful in contexts where the main thread in the program needs to continue in order to close out operations. This is used in the python API to properly (most of the time) handle Ctrl-C operations and terminate the co-simulation.
NOTE : the signal handlers use unsafe operations, so there is no guarantee they will work, or that they will work as expected. Testing indicates they work in most situations and improve operations where needed but it is not 100% reliable or safe code. They make use of atomic variables, mutexes, and other constructs that are not technically safe in signal handlers. The primary use case is program termination so the effects are minimized and they usually work, but the unsafe nature of them should be kept in mind.
/** Load a custom signal handler to execute prior to the abort signal handler. @details This function is not 100% reliable it will most likely work but uses some functions and techniques that are not 100% guaranteed to work in a signal handler and in worst case it could deadlock. That is somewhat unlikely given usage patterns but it is possible. The callback has signature HelicsBool(*handler)(int) and it will take the SIG_INT as an argument and return a boolean. If the boolean return value is HELICS_TRUE (or the callback is null) the default signal handler is run after the callback finishes; if it is HELICS_FALSE the default callback is not run and the default signal handler is executed. If the second argument is set to HELICS_TRUE the default signal handler will execute in a separate thread(this may be a bad idea). */ void helicsLoadSignalHandlerCallback(HelicsBool (*handler)(int), HelicsBool useSeparateThread);
It is also possible to insert a custom callback into the signal handler chain. Again this is not 100% reliable. But is useful for some language API’s that do other things to signals. This allows for inserting a custom operation that does some other cleanup. The callback has a helics_boolean return value. If the value is set to
HELICS_TRUE(or any positive value) then the normal or threaded Signal handler is called which aborts ongoing federations and exits. If it is set to
HELICS_FALSE then the default callback is not executed. If the
useSeparateThread call was set to true in the
helicsLoadSignalHandlerCallback method then the HELICS portion of the callback is executed in a new thread. The signal handler will return before this portion of the handler is completed to all control to return to the main program to complete operations. Once again these operations are not guaranteed to be safe in a signal handler. In most cases they work and can be useful for usability. And in most cases they are used when desiring to terminate a program so consequences are minimal.
Signal handlers in C++¶
Facilities for signal handling in C++ were not put in place since it is easy enough for a user to place their own handlers which would likely do a better job than any built in ones, so a default one was not put in place at present though may be at a later time.
Generating an error¶
A global error generated anywhere in a federation will terminate the co-simulation.
/** * generate a global error through a broker this will terminate the federation * * @param broker The broker to set the time barrier for. * @param errorCode the error code to associate with the global error * @param errorString an error message to associate with the error * @param[in,out] err An error object that will contain an error code and string if any error occurred during the execution of the function. */ void helicsBrokerGlobalError(HelicsBroker broker, int errorCode, const char *errorString, HelicsError* err); void helicsCoreGlobalError(HelicsCore core, int errorCode, const char* errorString, HelicsError* err); /** * Generate a global error from a federate. * * @details A global error halts the co-simulation completely. * * @param fed The federate to create an error in. * @param errorCode The integer code for the error. * @param errorString A string describing the error. */ HELICS_EXPORT void helicsFederateGlobalError(HelicsFederate fed, int errorCode, const char* errorString);
Corresponding functions are available in the C++ API as well. Any global error will cause a termination of the co-simulation.
Some modifying flags¶
HELICS_TERMINATE_ON_ERROR flag to true will escalate any local error into a global one and terminate the co-simulation. This includes any mismatched configuration or other local issues.
Generally it isn’t a wise idea to just terminate the co-simulation without letting everyone else know. If you control everything it probably works fine but as co-simulations get larger more care needs to be taken to prevent zombie processes and hung federates and brokers, which can cause issues on the next one. This is an evolving area of how best to handle terminating large co-simulations in abnormal conditions and hopefully the best practices will make it easier for users.